Julia将支持可组合的多线程并行机制

Chip designers are under so much pressure to deliver ever-faster CPUs that they’ll risk changing the meaning of your program, and possibly break it, in order to make it run faster

一个经典示例是并归排序，它将其输入分成两部分，并递归地对各部分排序。这两部分可以独立排序，自然带来了并行的可能。下面是代码:

import Base.Threads.@spawn

# sort the elements of `v` in place, from indices `lo` to `hi` inclusive
function psort!(v, lo::Int=1, hi::Int=length(v))
    if lo >= hi                       # 1 或 0 个元素，无需操作
        return v
    end
    if hi - lo < 100000               # 低于下限则串行运行
        sort!(view(v, lo:hi), alg = MergeSort)
        return v
    end

    mid = (lo+hi)>>>1                 # 找到中点

    half = @spawn psort!(v, lo, mid)  # 排序前半部分的任务将会与
    psort!(v, mid+1, hi)              # 当前排序后半部分的代码并行运行
                                      # 
    wait(half)                        # 等待前半部分完成

    temp = v[lo:mid]                  # 用于合并的空间

    i, k, j = 1, lo, mid+1            # 合并两个已排序的子数组
    @inbounds while k < j <= hi
        if v[j] < temp[i]
            v[k] = v[j]
            j += 1
        else
            v[k] = temp[i]
            i += 1
        end
        k += 1
    end
    @inbounds while k < j
        v[k] = temp[i]
        k += 1
        i += 1
    end

    return v
end

zzz6519003 / blog

Julia将支持可组合的多线程并行机制 #127